DEPR: to_datetime call in Series.combine_first (#62931)

Author: jbrockmendel

doc/source/whatsnew/v3.0.0.rst

@@ -737,6 +737,7 @@ Other Deprecations
 - Deprecated allowing ``fill_value`` that cannot be held in the original dtype (excepting NA values for integer and bool dtypes) in :meth:`Series.shift` and :meth:`DataFrame.shift` (:issue:`53802`)
 - Deprecated backward-compatibility behavior for :meth:`DataFrame.select_dtypes` matching "str" dtype when ``np.object_`` is specified (:issue:`61916`)
 - Deprecated option "future.no_silent_downcasting", as it is no longer used. In a future version accessing this option will raise (:issue:`59502`)
+- Deprecated silent casting of non-datetime 'other' to datetime in :meth:`Series.combine_first` (:issue:`62931`)
 - Deprecated slicing on a :class:`Series` or :class:`DataFrame` with a :class:`DatetimeIndex` using a ``datetime.date`` object, explicitly cast to :class:`Timestamp` instead (:issue:`35830`)
 - Deprecated the 'inplace' keyword from :meth:`Resampler.interpolate`, as passing ``True`` raises ``AttributeError`` (:issue:`58690`)
 

pandas/core/series.py

@@ -3267,6 +3267,16 @@ def combine_first(self, other) -> Series:
         if this.dtype.kind == "M" and other.dtype.kind != "M":
             # TODO: try to match resos?
             other = to_datetime(other)
+            warnings.warn(
+                # GH#62931
+                "Silently casting non-datetime 'other' to datetime in "
+                "Series.combine_first is deprecated and will be removed "
+                "in a future version. Explicitly cast before calling "
+                "combine_first instead.",
+                Pandas4Warning,
+                stacklevel=find_stack_level(),
+            )
+
         combined = concat([this, other])
         combined = combined.reindex(new_index)
         return combined.__finalize__(self, method="combine_first")

pandas/tests/series/methods/test_combine_first.py

@@ -2,6 +2,8 @@
 
 import numpy as np
 
+from pandas.errors import Pandas4Warning
+
 import pandas as pd
 from pandas import (
     Period,
@@ -75,9 +77,14 @@ def test_combine_first_dt64(self, unit):
         xp = to_datetime(Series(["2010", "2011"])).dt.as_unit(unit)
         tm.assert_series_equal(rs, xp)
 
+    def test_combine_first_dt64_casting_deprecation(self, unit):
+        # GH#62931
         s0 = to_datetime(Series(["2010", np.nan])).dt.as_unit(unit)
         s1 = Series([np.nan, "2011"])
-        rs = s0.combine_first(s1)
+
+        msg = "Silently casting non-datetime 'other' to datetime"
+        with tm.assert_produces_warning(Pandas4Warning, match=msg):
+            rs = s0.combine_first(s1)
 
         xp = Series([datetime(2010, 1, 1), "2011"], dtype=f"datetime64[{unit}]")